What is OpenAI and Anthropic?
Lead qualification AI quality depends on many factors. One of those factors is the model (brain) that’s used as the decision maker. Two very popular producers of these brains are OpenAI and Anthropic. Imagine OpenAI is Harvard and Anthropic is Yale. Both produce amazing brains, but they are trained in slightly different ways. Because they are trained in different ways, their responses are slightly different as well.
A High Level View of Differences
When you look at the differences between these companies, you’ll find that OpenAI is more concerned with pushing the envelop and training massive, robust models. Anthropic on the other hand has been more focused on training their models on quality-only data to improve the chat experience and safety of their product.
Everyone has heard of OpenAI, but surprisingly few have heard of Anthropic. This is likely because OpenAI came out guns blazing, releasing products seemingly without a care in the world about how they would be used. Anthropic, on the other hand, has been more calculated. They certainly fell behind OpenAI initially, but we are now seeing that their newer models like Opus produce better results in chat scenarios than OpenAI’s GPT-4 / Turbo models.
Example of How Each Performs
It’s most helpful to look at an example so we can see the differences first-hand. Both of these examples were run using the exact same CloseBot bot settings. Take a look at this sample exchange with Anthropic’s Opus (we use a mixture of Opus and Sonnet to save money for you). You’ll notice right away that Claude tends to be more wordy, but it does seem human-like in its responses. It tends to not feel as much like an assistant, and feels more like it understands the nuance in speech and can respond with correct emotion.
Anthropic Example (Opus)
OpenAI’s models do tend to generate shorter responses. However, we’ve found that the responses also tend to feel more robotic (even when using the exact same prompting). In this example, the AI doesn’t make me feel validated. I don’t feel that it cares about my predicament like Anthropic did above.
OpenAI Example (GPT-4 Turbo)
Cost
Of course, there’s more to lead qualification AI than quality. We also have to consider cost. When using your own API key (no up-charge from CloseBot), the average cost per message is as follows:
Even though Anthropic’s Opus is considerably more expensive than GPT-4 Turbo, we are able to use Anthropic’s Sonnet and Haiku models for some of the back end multi-agent work. These models are cheaper than GPT-4 Turbo and it’s this model shuffling that makes Anthropic the cheaper choice. OpenAI’s cheaper model GPT-3.5 Turbo is unable to handle most of our multi-agent work, so we are forced to use their more expensive model(s) for these stages.
Summary
CloseBot has access to both OpenAI and Anthropic. It’s important to use a lead qualification AI product that makes use of both because it proves they can pivot depending on which AI company is leading the pack, and it allows other options if one of the companies experiences downtime! Check out our no-prompt lead qualification AI builder for free and experience the power yourself!
Comments are closed.